Investigating The Market’s “Nightmare Scenario”

Right, so one thing people are becoming increasingly concerned about is the extent to which modern market “innovations” could end up backfiring in spectacular fashion in the event the current low vol, BTFD regime were to suddenly slam into reverse.

I wrote about this a couple of days ago and this is probably a hyperbolic, overwrought assessment, but it captures the essence of the potential problem:

…it’s looking more and more like this won’t play out over six or eight months like it did during the crisis. Instead, it’s looking like this could play out over six or eight days.

Again, probably exaggerated, but at issue is the notion that between things like VIX ETPs and the proliferation of programmatic strats, we’ve created a situation where a whole bunch of potentially destructive feedback loops could conspire to create one giant self-fulling prophecy.

One wonders, for instance, what would happen if rebalancing by VIX ETPs after a sudden vol spike exacerbated said vol spike which in turn forced vol control funds and CTAs to deleverage all at once. And then there’s the risk parity component.

Obviously this is Marko Kolanovic’s wheelhouse and those interested in the dangers posed by these kinds of systematic, model-driven unwinds are encouraged to read the following posts:

Relatedly, it’s worth checking out this from Kevin Muir:

Guest Post: Axe Would Hate Risk Parity

At the end of the day, the risk that the kind of nightmare scenario described above actually comes calling is impossible to pin down and the damage it would cause is equally difficult to quantify. But that doesn’t mean people aren’t trying.

Because it seems like this type of cascading collapse gets more likely with each passing day that volatility remains glued to the proverbial flatline, we thought it was worth presenting the following excerpts from DB and then from BofAML.

In the context of everything said above, you’re encouraged to draw your own conclusions because after all, what’s the point of presenting a bunch of analysis only to close by telling you that no matter what anyone says, it’s all going to fall apart?

Via DB (from a note dated last week)

VIX ETPs’ market impact growing; vol control fund rebalancing slowed by low realized vol

Vega that VIX ETP providers would have to buy if volatility spiked to hedge exposure is near all-time highs this month, as the vega in short VIX ETPs has grown. Vega to hedge on a spike has continued its steady rise since the vol increase around US elections and sits close to $90mm.

Yet in terms of overall impact in the market, the vega to sell sits at just ~40% of the overall VIX futures market, making it less of a risk than it was at times last year. However, it is uncertain how liquid the VIX futures market would be after that kind of vol spike. The large amount of vega to buy may be hard for the market to absorb in some stress events, which may cause further exacerbation of a vol spike.

Vol control funds that track a certain level of realized vol have not seen significant selling since the volatility around Brexit; even the US election only saw slight drawdowns. There has been very little deleveraging due to the low levels of realized vol. The selling around the US elections were the only real deleveraging since last June, and funds reallocated within a few weeks.

These funds could theoretically be even more allocated if allowed to take on leverage, so on a quick spike in volatility now, some funds would just get near their target level and not above it. It will take a longer period of high vol to cause large-scale selling in vol control funds.

Some of the vol control funds that track an implied volatility metric saw some indicators for selling around the first round of the French elections, but likely not in any significant size. The VIX futures curve inverted, but didn’t increase to levels that would indicate a true risk off.

CTAs appear to be signaling “risk on” trading tendencies, as well as a preference for equities ex-US

CTA exposure to large cap US equities has remained near 1 year lows, while the CTA betas of equities ex-US and commodities have grown over the past year. Exposure to MSCI EM and EAFE have been higher than to the US recently. The stronger negative beta to treasuries over the last year may be another indicator of “risk on” from the CTA community.

CTAs have less of a relationship with the dollar over the last week/month. The US dollar has been a focus of markets over the last year as both the new administration and the Fed have discussed changes to interest and exchange rates and the currency was near 12 year highs. With less interest in USD and SPX, CTA managers seem to be less US focused than they were in 2016 and early 2017.

Running a multiple regression on the HFRX Global Hedge Fund Index using global equity markets, commodities, oil, gold, USD, and Treasuries allows us to study current beta exposures for CTA and macro funds. The multiple regression beta proxies the CTA exposure to each asset class, independent of the movement of other assets.

CTA reallocation can be a magnifier of market volatility along the same vein as the other feedback loops. It is not only the overall beta number that is important, which tells us their allocation relative to other assets,but how that regression beta changes over time, and the rate at which it changes. This can give insight into how quickly CTAs are allocating to or leaving certain positions, which can lead to market volatility for the assets.

Via BofAML (from a noted dated late last month)

Higher stocks on ultra-low vol leave CTAs long equities

Based on a recent analysis of price trends across asset classes we noted that CTAs likely had high exposure to equities. Despite seesawed equity markets over the last few weeks, our models still indicate that CTAs have outsized positions in global equities. In fact, the current allocation to global equities from our bottom-up model of CTA positioning is just off its highest levels since at least the start of 2015 (Chart 1).

Higher equity allocations mean CTAs have an increased beta to equities. A consequence of extreme levels of equity positioning for CTAs is an increased beta to equity market moves and specifically during equity market sell-offs. For CTAs who are risk controlled, which we think most are, in the current environment it could take a relatively smaller equity market decline to trigger rules-based selling pressure. However, as different CTAs have varying sets of rules and risk control mechanisms, the selling pressure may not necessarily occur simultaneously or even within the same day.

Estimating CTA flows in an equity risk-off event

A simple analysis of price trends and volatility makes a reasonable case for multi-year high equity allocations for CTAs. If the combination of higher stocks on ultra-low vol is indeed putting CTAs’ equity positions at elevated risks of sudden deleveraging, then it is important to estimate the contribution of potential selling pressure versus liquidity of the markets in which they trade, particularly in a risk-off event. However, estimating CTA equity selling pressure is relatively more difficult than rationalizing their outsized positioning as, for example, key questions surround (1) what percentage of total CTA assets are in explicit rules-based, trend following strategies and importantly (2) how diverse the models are for rules-based, trend following CTAs.

According to BarclayHedge, AUM within the CTA industry through the end of 2016 stood at approximately $250bn. In our recent QIS Panorama we showed that the performance of the Altegris 40 Index (an asset-weighted benchmark CTA index priced monthly) could be well-explained by cross-asset, risk controlled trend following strategies. This benchmark currently accounts for $106bn in assets (~40% of the total CTA AUM) and therefore we conservatively estimate the potential range of assets in rules-based trend following strategies as between ~$100bn and ~$250bn.

In order to understand positioning and potential flows from CTAs we built a bottom-up, rules-based, trend following CTA model that allocates to multiple cross-asset futures and we believe well-represents the universe of CTA investments on average. Our model has strong correlation and beta near one to daily returns of the SG CTA Index which therefore allows us to draw reasonable conclusions on CTA allocations and trading flows as a function of market dynamics.

The current allocation to global equities from our bottom-up CTA model is about 70% of total assets (Chart 1). This equates to somewhere between $70bn and $175bn of global equity exposure for trend following CTAs. Global equity allocations in our model are represented by investments in the three most liquid futures contracts in each of the major regions (US: SPX, NDX, RTY; Europe: SX5E, DAX, UKX, & Asia: NKY, HSI, KOSPI2)

As mentioned earlier, CTAs’ current global equity positioning is historically high and therefore it could take a smaller equity market decline to trigger rules-based selling pressure. Our model global equity allocations and required 1-day market moves to force an entire unwind are shown in Table 1. While some of the 1-day declines seem feasible (e.g. SPX ~1.5%), others appear less attainable (e.g. RTY ~10%). Even though some moves that would drive a large model-driven unwind seem possible, given the extreme lows in equity volatility currently, many are high relative to current volatility (e.g. ~1.5% SPX 1-day decline would be near a 3.6-sigma move currently).

Successive shocks likely needed for full equity unwind but liquidity also higher. Based on history, there is a low probability that markets could move enough in one day to trigger a full unwind of current record CTA equity positions. However, successive ‘high-sigma’ declines in today’s environment could happen with greater likelihood. By successive high-sigma declines, we mean multiple days of down moves that are high relative to prevailing volatility. In Table 2 are three 5-day periods since 2006 in which global equity markets saw successive high-sigma declines that could in the current environment trigger a complete unwind of near record CTA equity positions.

Should equity markets decline in similar fashion to the one week periods from late Feb- 07, the Aug-11 US credit downgrade, or the Aug-15 sell-off then according to our model CTAs would unwind mostly all of their global equity positions to limit losses. Based on our estimates, that would be between $70bn and $175bn in global equity futures selling pressure which as a percentage of 3-month median daily global equity futures notional volume is between 20% and 55%. This amount of impact in one day may seem large but does not take into account two important factors. First, in stress events similar to Feb-07, Aug-11, or Aug-15 it’s reasonable to expect a substantial increase in equity index futures volume versus prevailing median levels. Second, in order for the entire unwind to happen within one day, we would need (1) global equity markets to see a never-before seen shock in sigma terms and (2) the entirety of CTA assets to be in pure trend following programs with an extremely similar set of rules to limit drawdowns which we do not think is the case.

However, it is more reasonable to consider a situation in which CTAs collectively unwind their global equity positions over multiple days of consecutive high-sigma declines. Through each of the stress events in Table 2 we measured the increase in the respective equity future indices’ notional volume through the 5-day decline versus the median 5-day notional volume over the three months prior. On average, most major global equity index futures saw a doubling in prevailing volume in stress events (Chart 7). Therefore, given current 3-month median weekly notional futures volume on the nine global equity indices of about $1.6 trillion, if we see an event large enough to trigger a model driven unwind of CTA equity positions then we estimate global equity index futures volume rising to $3.2 trillion (Chart 8). In this extreme, if the entirety of our highest estimate for CTA’s total equity allocations (~$175bn) is unwound within a week of successive high-sigma equity market declines, then it would equate to only ~6% of the expected volume.

The estimated equity selling flows from CTAs in high-sigma events could have an impact on markets but it is quite difficult to suggest they would be the dominant driver in a severe shock. While not being the primary seller, conceptually CTAs can have the scope to add convexity to a fundamentally-driven sell-off as they could become a nondiscretionary seller in times when the market is already declining. In this case, the potential flows may become more meaningful depending on how uniform CTA models are, how quickly they react to changing price trends, and as well what portion of CTA assets are represented by rules-based trend-following strategies.

Fear of CTA flows may be the greater threat

When fears of CTAs driving the market lower become a self-fulfilling prophecy. While our expectations of potential CTA equity deleveraging flows may not dominate volumes in isolation, a remaining unknown is the additional selling pressure from investors fearful of these model driven flows. But even if we were to see volumes increase up to 3x daily levels (vs. 1.5x from CTAs alone), the biggest one-day sigma decline since 2000 would imply only a 2% decline in global equities in a single day. For an even broader perspective, since 1928 the largest one-day S&P 500 sigma decline observed was 11x which off today’s MSCI vol levels equates to only a 3.5% down day.

It’s important to also keep in mind that CTAs will only react to downward price action and that they will not instigate it. But if a sell-off comes in response to no immediate or obviously accountable macro catalyst, then some may look to the role of CTAs in trying to explain the market decline. The fear of CTA’s rules-based, nondiscretionary selling flows in stress periods may cause other more fundamental and discretionary managers to also unwind which could then potentially create a negative feedback loop of successive declines in equity markets.

Further, as we’ve noted in the past, equity futures market liquidity is not only measured by total volume but also market depth which in recent risk-off events has declined in times of vol shocks. The greater the selling pressure that is concentrated in a shorter time period against falling market depth (when liquidity is needed most) could also further exacerbate market shocks.

However, the idea that CTAs can cause another ’87-style crash in our opinion is likely too extreme. First, on 19-Oct-87, the S&P 500 declined nearly 21% in one day which given today’s vol levels would be a 56-sigma event and nearly 5x higher any daily sigma event the S&P 500 has seen since 1928. To observe a historically significant one-day decline, volatility would likely need to be much higher. However, if volatility were indeed higher, then equity positioning in CTAs should be lower and therefore also the amounts they can deleverage. Lastly, circuit breakers that halt trading after sufficient declines which were introduced in response to events like 1987 should also help reduce the likelihood of similar shocks occurring again.

Risk parity & vol control, the one trillion dollar question

Model driven equity selling pressure via risk parity and equity vol control strategies is often also considered alongside that of CTAs as they all (1) use rules-based models that can at times make them price-insensitive buyers or sellers, (2) typically increase leverage when volatility is lower, and (3) can deleverage in response to a shock from low vol levels.

As with CTAs, estimating potential equity selling pressure from risk parity and equity vol control funds requires first knowing how much of their assets are purely rules-based and second modelling how they could operate in a stress event. However, relative to CTAs there is much less transparency on the total size of assets in risk parity and equity vol control strategies let alone the subset of which is completely rules-based.

Regardless, we can estimate how rules-based risk parity and equity vol control funds would respond in an Aug-15 like stress event where equities were subject to multiple days of large declines and bonds did not offer sufficient diversification. Our models estimate that in shocks similar to Aug-15, some rules-based risk parity strategies could sell equities in the amount of 25% of their total assets under management. In addition, we estimate that some rules-based equity vol control strategies could sell equities in the amount of 120% of their total AUM.

However, due to model diversity, we would not expect rules-based risk parity and equity vol control funds to all unwind within a day. Rather, in an event similar to Aug-15 in which equity markets were subject to consecutive days of large declines, we expect these funds to deleverage over the course of a few days or a week depending on how stress unfolds. In this case, as we discussed earlier, equity index futures volumes over the week could be expected to double from current levels to an amount north of $3tn.

The key question is how big would these quant fund assets need to be to represent a significant portion of futures volumes? When considering that unwinds during stress could take upwards of a week, the asset size required to be invested such that they would represent even 20% of total weekly volumes seems far larger than what we believe exists in the market.

To account for 20% of expected weekly volume of $3tn, equity selling flows would need to equal $600bn. Earlier in this piece we accounted for a worst case $175bn equity selling pressure from CTAs (which we think is an overestimate). To get to $600bn in total selling pressure would require rules-based risk parity to have upwards of $500bn in assets with an additional $250bn in equity vol control. Therefore, it would take $1 trillion (far higher than any estimates we have seen in the market) in completely rulesbased assets across CTAs, risk parity, and equity vol control funds for the trio to account for even 20% of expected weekly equity index futures volume in severe stress. This would be in one in which equity vol spikes from ultra-low levels, markets decline for multiple days in a row, and in which bonds fail to sufficiently diversify equities. However, this amount of assets is far beyond any reasonable estimate in our view.

Again, it’s important to recognize that markets could react negatively to model driven flows in today’s fragile environment, particularly as there is little transparency about how they operate or their potential size. However, appreciating the likely limits of their impact is key to better navigating stress events, and identifying how to take advantage of a potential overreaction from investors scared of what they don’t understand.